The Agent Skills Directory

[DYNAMIC_CONTEXT_INJECTION]: The skill uses dynamic context commands (prefixed with !) to harvest repository topology, such as branch information, manifest existence, and service enumeration from docker-compose.yml. This data is used to ground the agent's understanding of the system.
[COMMAND_EXECUTION]: The skill executes several shell commands and custom scripts to analyze the project, including git, docker compose, terraform show, and jq. These operations are limited to architectural discovery and state management.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted external data in the form of stack traces or incident logs (e.g., Sentry JSON, OTel spans) in trace mode. While this constitutes an attack surface for indirect prompt injection, the skill's capabilities are focused on HTML generation and local state management.
Ingestion points: Trace files provided via $ARGUMENTS (e.g., sentry-event.json).
Boundary markers: None explicitly defined for trace parsing.
Capability inventory: Bash (git, ls, cat, jq, docker, kubectl, terraform, python3), Write, Glob, Grep.
Sanitization: The skill uses a custom HTML validator (validate_html.py) to ensure accessibility and check for unauthorized external scripts, but does not explicitly sanitize natural language content inside traces.
[EXTERNAL_DOWNLOADS]: The generated HTML templates reference Google Fonts and the Mermaid.js library from cdn.jsdelivr.net. These are well-known, trusted services used for the intended purpose of rendering diagrams.

engineer-design-diagram